• Angelici, G. L., J. W. Skiles, and L. Z. Popovici, 1991: OTTER pilot land data system, vol. 1, version 1: Satellite, aircraft and ground measurements. CD-ROM, NASA. [Available from Oak Ridge National Laboratory DAAC, P.O. Box 2008, MS 6407, Oak Ridge, TN 37831.].

  • Baldwin, D. G., and W. J. Emery, 1993: Systematized approach to AVHRR navigation. Ann. Glaciol.,17, 414–420.

    • Crossref
    • Export Citation
  • Bates, D. M., and D. G. Watts, 1988: Nonlinear Regression Analysis and its Applications. John Wiley and Sons, 384 pp.

    • Crossref
    • Export Citation
  • Betts, A. K., P. Viterbo, A. Beljaars, H.-L. Pan, S.-Y. Hong, M. Goulden, and S. Wofsy, 1998: Evaluation of land-surface interaction in ECMWF and NCEP/NCAR reanalysis models over grassland (FIFE) and boreal forest (BOREAS). J. Geophys. Res.,103, 23 079–23 085.

    • Crossref
    • Export Citation
  • Bounoua, L., G. J. Collatz, S. O. Los, P. J. Sellers, D. A. Dazlich, C. J. Tucker, and D. A. Randall, 2000: Sensitivity of climate to changes in NDVI. J. Climate, in press.

    • Crossref
    • Export Citation
  • Burgess, A. W., P. Lewis, and J. P. A. L. Muller, 1995: Topographic effects in AVHRR NDVI data. Remote Sens. Environ.,54, 223–232.

    • Crossref
    • Export Citation
  • Chen, J. M., 1996: Canopy architecture and remote sensing of the fraction of photosynthetically active radiation absorbed by boreal conifer forests. IEEE Trans. Geosci. Remote Sens.,34, 1353–1368.

    • Crossref
    • Export Citation
  • ——, and J. Cihlar, 1996: Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sens. Environ.,55, 153–162.

    • Crossref
    • Export Citation
  • ——, P. M. Rich, S. T. Gower, J. M. Norman, and S. Plummer, 1997:Leaf area index of boreal forests: Theory, techniques, and measurements. J. Geophys. Res.,102, 29 429–29 444.

  • Choudhury, B. J., 1987: Relationships between vegetation indices, radiation, absorption and net photosynthesis evaluated by a sensitivity analysis. Remote Sens. Environ.,22, 209–233.

    • Crossref
    • Export Citation
  • Collatz, G. J., J. A. Berry, and J. S. Clark, 1998: The effect of climate and atmospheric CO2 partial pressure on the global distribution of C4 grasses: Present, past, and future. Oecologia,114, 441–454.

    • Crossref
    • Export Citation
  • Cramer, W., D. W. Kicklighter, A. Bondeau, B. Moore, C. Churkina, B. Nemry, A. Ruimy, and A. Schloss, 1999: Comparing global models of terrestrial net primary productivity (NPP): Overview and key results. Global Change Biol.,5, 1–15.

    • Crossref
    • Export Citation
  • Deering, D. W., and P. Leone, 1986: A sphere-scanning radiometer for rapid directional measurements of sky and ground radiance. Remote Sens. Environ.,19, 1–24.

    • Crossref
    • Export Citation
  • DeFries, R. S., and J. R. G. Townshend, 1994: NDVI-derived landcover classifications at global scale. Int. J. Remote Sens.,15, 3567–3586.

    • Crossref
    • Export Citation
  • ——, ——, and S. O. Los, 1997: Scaling landcover heterogeneity for global atmosphere–biosphere models. Scale in Remote Sensing and GIS, D. A. Quattrochi and M. F. Goodchild, Eds., Lewis Publishers, 231–246.

  • ——, M. Hansen, J. R. G. Townshend, and R. Sohlberg, 1998: Global landcover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers. Int. J. Remote Sens.,19, 3141–3168.

  • Denning, A. S., I. Y. Fung, and D. A. Randall, 1995: Latitudinal gradient of atmospheric CO2 due to seasonal exchange with land biota. Nature,376, 240–243.

    • Crossref
    • Export Citation
  • Dirmeyer, P. A., A. J. Dolman, and N. Sato, 1999: The Global Soil Wetness Project: A pilot project for global land surface modeling and validation. Bull. Amer. Meteor. Soc.,80, 851–878.

    • Crossref
    • Export Citation
  • Draper, N. R., and H. Smith, 1981: Applied Regression Analysis. John Wiley and Sons, 709 pp.

  • Eischeid, J. K., H. F. Diaz, R. S. Bradley, and P. D. Jones, Eds., 1991:A comprehensive precipitation dataset for global land areas. U.S. Dept. of Energy Carbon Dioxide Research Program Rep. TR051, 82 pp. [Available from Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, TN 37831.].

  • Field, C. B., J. T. Randerson, and C. M. Malmström, 1995: Global net primary production: Combining ecology and remote sensing. Remote Sens. Environ.,51, 74–88.

    • Crossref
    • Export Citation
  • Fournier, R. A., P. M. Rich, and R. Landry, 1997: Hierarchical characterization of canopy architecture for boreal forest. J. Geophys. Res.,102, 29 445–29 454.

    • Crossref
    • Export Citation
  • Fung, I. Y., C. J. Tucker, and K. C. Prentice, 1987: Application of Advanced Very High Resolution Radiometer vegetation index to study atmosphere–biosphere exchange of CO2. J. Geophys. Res.,92, 2999–3015.

    • Crossref
    • Export Citation
  • ——, and Coauthors, 1997: Carbon 13 exchanges between the atmosphere and biosphere. Global Biochem. Cycles,11, 507–533.

    • Crossref
    • Export Citation
  • Goward, S. N., and K. F. Huemmrich, 1992: Vegetation canopy PAR absorbtance and the normalized difference vegetation index: An assessment using the SAIL model. Remote Sens. Environ.,39, 119–140.

    • Crossref
    • Export Citation
  • ——, B. Markham, D. G. Dye, W. Dulaney, and J. Yang, 1991: Normalized difference vegetation index measurements from the Advanced Very High Resolution Radiometer. Remote Sens. Environ.,35, 257–277.

    • Crossref
    • Export Citation
  • Gutman, G. G., 1991: Vegetation indices from AVHRR: An update and future prospects. Remote Sens. Environ.,35, 121–136.

    • Crossref
    • Export Citation
  • ——, and A. Ignatov, 1995: Global land monitoring from AVHRR: Potential and limitations. Int. J. Remote Sens.,16, 2301–2309.

    • Crossref
    • Export Citation
  • Hall, F. G., K. F. Huemmrich, S. J. Goetz, P. J. Sellers, and J. E. Nickeson, 1992: Satellite remote sensing of surface energy balance: Success, failures, and unresolved issues in FIFE. J. Geophys. Res.,97, 19 061–19 089.

    • Crossref
    • Export Citation
  • Hanan, N. P., A. Bégué, and S. D. Prince, 1997: Errors in remote sensing of intercepted photosynthetically active radiation: An example from HAPEX—Sahel. J. Hydrol.,188–189, 676–696.

    • Crossref
    • Export Citation
  • Haxeltine, A., and I. C. Prentice, 1996: BIOME3: An equilibrium terrestrial biosphere model based on ecophysiological constraints, resource availability and competition among plant functional types. Global Biogeochem. Cycles,10, 693–709.

  • Holben, B. N., 1986: Characteristics of maximum-value composite images for temporal AVHRR data. Int. J. Remote Sens.,7, 1435–1445.

    • Crossref
    • Export Citation
  • James, M. E., and S. N. V. Kalluri, 1994: The Pathfinder AVHRR land dataset: An improved coarse resolution dataset for terrestrial monitoring. Int. J. Remote Sens.,15, 3347–3363.

    • Crossref
    • Export Citation
  • Jones, P. D., S. C. B. Raper, B. Santer, B. S. G. Cherry, C. Goodess, P. M. Kelly, T. M. L. Wigley, R. S. Bradley, and H. F. Diaz, Eds., 1985: A grid point surface air temperature dataset for the Northern Hemisphere. U.S. Dept. of Energy Carbon Dioxide Research Division Tech. Rep. TR022, 251 pp. [Available from Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, TN 37831.].

  • Justice, C. O., T. F. Eck, D. Tanré, and B. N. Holben, 1991: The effect of water vapor on the normalized difference vegetation index for the Sahelian region from NOAA-AVHRR data. Int. J. Remote Sens.,12, 1165–1187.

    • Crossref
    • Export Citation
  • Keeling, C. D., T. P. Whorf, M. Wahlen, and J. van der Plicht, 1995:Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980. Nature,375, 666–670.

    • Crossref
    • Export Citation
  • Los, S. O., 1998a: Estimation of the ratio of sensor degradation between NOAA-AVHRR channels 1 and 2 from monthly NDVI composites. IEEE Trans. Geosci. Remote Sens.,36, 206–213.

    • Crossref
    • Export Citation
  • ——, 1998b: Linkages between global vegetation and climate: An analysis based on NOAA-Advanced Very High Resolution Radiometer data. Ph.D. dissertation, Vrije Universiteit Amsterdam, 199 pp. [Also available as NASA Rep. GSFC/CR-1998-206852;NASA-CASI, Hanover, MD 21076-1320.].

  • ——, C. O. Justice, and C. J. Tucker, 1994: A global 1° by 1° NDVI dataset for climate studies derived from the GIMMS continental NDVI. Int. J. Remote Sens.,15, 3493–3518.

    • Crossref
    • Export Citation
  • Loveland, T. R., and A. S. Belward, 1997: The IGBP-DIS global 1 km landcover dataset, DISCover: First results. Int. J. Remote Sens.,18, 3289–3295.

    • Crossref
    • Export Citation
  • Malmström, C. M., M. V. Thompson, G. Juday, S. O. Los, J. T. Randerson, and C. B. Field, 1997: Interannual variation in global-scale primary production: Testing model estimates. Global Biochem. Cycles,11, 367–392.

    • Crossref
    • Export Citation
  • Myneni, R. B., S. O. Los, and C. J. Tucker, 1995: Satellite-based identification of linked vegetation index and sea surface temperature anomaly areas from 1982–1990 for Africa, Australia and South America. Geophys. Res. Lett.,23, 729–732.

    • Crossref
    • Export Citation
  • ——, R. R. Nemani, and S. W. Running, 1997a: Estimation of global leaf area index and absorbed PAR using radiative transfer models. IEEE Trans. Geosci. Remote Sens.,35, 1380–1393.

    • Crossref
    • Export Citation
  • ——, C. D. Keeling, C. J. Tucker, G. Asrar, and R. R. Nemani, 1997b:Increased plant growth in the northern high latitudes from 1981 to 1991. Nature,386, 698–702.

    • Crossref
    • Export Citation
  • Nemani, R. R., S. W. Running, R. A. Pielke, and T. N. Chase, 1996:Global vegetation cover changes from coarse resolution satellite data. J. Geophys. Res.,101, 7157–7162.

    • Crossref
    • Export Citation
  • Potter, C. S., J. T. Randerson, C. B. Field, P. A. Matson, P. M. Vitousek, H. A. Mooney, and S. A. Klooster, 1993: Terrestial ecosystem production: A process model based on global satellite and surface data. Global Geochem. Cycles,7, 811–841.

    • Crossref
    • Export Citation
  • Price, J. C., 1987: Calibration of satellite radiometers and the comparison of vegetation indices. Remote Sens. Environ.,21, 15–27.

    • Crossref
    • Export Citation
  • Privette, J. L., C. Fowler, G. A. Wick, D. Baldwin, and W. J. Emery, 1995: Effects of orbital drift on Advanced Very High Resolution Radiometer products: Normalized difference vegetation index and sea surface temperature. Remote Sens. Environ.,53, 164–171.

    • Crossref
    • Export Citation
  • Randall, D. A., and Coauthors, 1996: A revised land-surface parameterization (SiB2) for GCMs. Part III: The greening of the Colorado State University general circulation model. J. Climate,9, 738–763.

    • Crossref
    • Export Citation
  • Randerson, J. T., M. V. Thompson, C. M. Malmstrom, C. B. Field, and I. Y. Fung, 1996: Substrate limitations for heterotrophs: Implications for models that estimate the seasonal cycle of atmospheric CO2. Global Biochem. Cycles,10, 585–602.

  • Reyment, R., and K. G. Jöreskog, 1993: Applied Factor Analysis in the Natural Sciences. Cambridge University Press, 371 pp.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev.,115, 1606–1626.

    • Crossref
    • Export Citation
  • Roujean, J. L., M. Leroy, and P. Y. Deschamps, 1992: A bidirectional reflectance model of the earth’s surface for the correction of remote sensing data. J. Geophys. Res.,97, 20 455–20 468.

    • Crossref
    • Export Citation
  • Ruimy, A., B. Saugier, and G. Dedieu, 1994: Methodology for the estimation of terrestrial net primary production from remotely sensed data. J. Geophys. Res.,99, 5263–5283.

    • Crossref
    • Export Citation
  • Sellers, P. J., 1985: Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens.,6, 1335–1372.

    • Crossref
    • Export Citation
  • ——, J. A. Berry, G. J. Collatz, C. B. Field, and F. G. Hall, 1992: Canopy reflectance, photosynthesis and transpiration, III. A re-analysis using improved leaf models and a new canopy integration scheme. Remote Sens. Environ.,42, 187–216.

    • Crossref
    • Export Citation
  • ——, D. A. Randall, G. J. Collatz, J. A. Berry, C. B. Field, D. A. Dazlich, C. Zhang, and L. Bounoua, 1996a: A revised land-surface parameterization (SiB2) for GCMs. Part I: Model formulation. J. Climate,9, 676–705.

    • Crossref
    • Export Citation
  • ——, S. O. Los, C. J. Tucker, C. O. Justice, D. A. Dazlich, G. J. Collatz, and D. A. Randall, 1996b: A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate,9, 706–737.

    • Crossref
    • Export Citation
  • ——, and Coauthors, 1996c: The ISLSCP Initiative I global datasets:Surface boundary conditions and atmospheric forcings for land–atmosphere studies. Bull. Amer. Meteor. Soc.,77, 1987–2005.

    • Crossref
    • Export Citation
  • Singh, S. M., 1988: Simulation of solar zenith angle effect on global vegetation index (GVI) data. Int. J. Remote Sens.,9, 237–248.

    • Crossref
    • Export Citation
  • Strebel, D. E., D. R. Landis, K. F. Huemmrich, and B. W. Meeson, 1994: Collected data of the First ISLSCP Field Experiment, vol. 1: Surface observations and non-image data sets. CD-ROM, NASA. [Available from Oak Ridge National Laboratory P.O. Box 2008, DAAC, MS 6407, Oak Ridge, TN 37831.].

  • Tucker, C. J., and P. J. Sellers, 1986: Satellite remote sensing of primary production. Int. J. Remote Sens.,7, 1395–1416.

    • Crossref
    • Export Citation
  • ——, H. E. Dregne, and W. W. Newcomb, 1994: AVHRR datasets for determination of desert spatial extent. Int. J. Remote Sens.,15, 3547–3565.

    • Crossref
    • Export Citation
  • Vermote, E. F., N. El Saleous, Y. J. Kaufman, and E. Dutton, 1997a:Data preprocessing: Stratospheric aerosol perturbing effect on the remote sensing of vegetation: Correction method for the composite NDVI after the Pinatubo eruption. Remote Sens. Rev.,15, 7–21.

    • Crossref
    • Export Citation
  • ——, D. Tanré, J. L. Deuzé, M. Herman, and J. J. Morcette, 1997b:Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An overview. IEEE Trans. Geosci. Remote Sens.,35, 675–686.

    • Crossref
    • Export Citation
  • View in gallery

    Interferences in NDVI data. (a) Variations in NDVI over the Saharan desert that result from calibration differences and sensitivity changes in AVHRRs on board NOAA-7, -9, and -11. (b) Effects of optical thickness from El Chichón aerosols on average monthly NDVI values >0.5. (c) Temporal changes and missing data from clouds in an NDVI time series over a tropical forest with otherwise low temporal variation. (d) Decreased NDVI values at the start of the rainy season in the Sahel. Deviations are related to high atmospheric water vapor in the ITCZ (Justice et al. 1991). Dotted lines indicate estimates of NDVI with water vapor effects removed. (e) (top line) Viewing angle–dependent variations in the Roujean et al. (1992) model fitted through NDVI data from FIFE (Deering and Leone 1986). (bottom line) Viewing angle–dependent variations in the same NDVI data at the top of the atmosphere. Viewing angle variations were simulated with 6S (Vermote et al. 1997b) for a midlatitude continental atmosphere with fairly high aerosol concentration. The ground-measured NDVI data and top-of-the-atmosphere data have an inverse relationship. A small hot-spot effect exists at 30° off nadir in the backscatter direction. (f) The effect of solar zenith angle on NDVI (ΔNDVI/Δθ) estimated from the global dataset (line marked 1) compared with ground-measured NDVI (top curve) and simulated top-of-the-atmosphere NDVI for clear (optical thickness = 0.1) and hazy (optical thickness = 0.5) atmospheres. The solar zenith angle effect for line “1” is within the simulations for clear and hazy atmospheres.

  • View in gallery

    Outline of FASIR corrections: (a) Original NDVI time series with outliers (two lowest NDVI values). (b) NDVI time series after Fourier adjustment (FA of FASIR; open circles are new estimates). (c) Relationship between NDVI and solar zenith angle for fully green canopies. For a given solar zenith angle, its effect is assumed to be linear with NDVI. (d) Interpolation of missing data during the winter. (e) Reconstruction of NDVI time series over tropical forests. The seasonality of tropical forests is removed.

  • View in gallery

    Evaluation of the Fourier adjustment. The months of Jan, Apr, Jul, and Oct were removed from a sequence of 12 months and their NDVI values were estimated with the Fourier adjustment. The figure shows the frequency distributions of the difference between the estimated and observed NDVI data with several of the statistics of the distributions (minimum, 5th percentile, median; 95th percentile, maximum). No comparison was made for missing values in the observed NDVI data.

  • View in gallery

    Derivation of FPAR absorbed by the vegetation canopy and leaf area index from NDVI. (a) Calculation of 2d and 98th percentile values from NDVI frequency distributions per vegetation class. The 98th percentile values were derived from tall vegetation class (assumed to be fully green) and the 2d percentile values were derived from desert and bare soil and shrub classes (assumed to be bare). (b) Calculation of FPAR from NDVI. FPAR is estimated as the mean of a linear relationship between FPAR and NDVI and a linear relationship between FPAR and SR = (1 + NDVI)/(1 − NDVI). The 98th percentile NDVI and SR values correspond to FPAR = 0.95 and the 2d percentile NDVI and SR values correspond to FPAR = 0.001. (c) Vegetation cover fraction fV is estimated as the maximum FPAR over the year divided by the maximum FPAR for the class (0.95). (d) Calculation of leaf area index from FPAR. An exponential relationship is used for vegetation within the vegetation cover fraction.

  • View in gallery

    Evaluation of NDVI FPAR relationships with data from FIFE (Strebel et al. 1994; Hall et al. 1992), OTTER (Angelici et al. 1991), BOREAS [data provided by C. Walthall, P. Rich, K. F. Huemmrich, and the BOREAS information system; see also Fournier et al. (1997), Chen et al. (1997), and Chen (1996)], and HAPEX—Sahel (Hanan et al. 1997). (a) Measured FPAR vs FPAR estimated with the NDVI–FPAR model. (b) Measured FPAR vs FPAR estimated with the NDVI–SR model. (c) Measured FPAR vs mean estimated FPAR from the NDVI–FPAR and SR–FPAR model (see text for discussion).

  • View in gallery

    Relationship between NDVI corrected for Rayleigh scattering and ozone absorption and uncorrected NDVI. Data were taken from a N–S transect through Africa from the Sahara, through the equatorial forest, and into southern Africa. Data are from NOAA-9 maximum NDVI composites of Mar 1995.

  • View in gallery

    Evaluation of FPAR–leaf area index relationships. The symbols indicate ground-measured FPAR vs leaf area index. The solid lines show the exponential model and the dotted lines the linear model used to estimate leaf area index from FPAR. (a) Total (from both live and dead vegetation) FPAR vs total leaf area index measured at the FIFE field campaign (Strebel et al. 1994). (b) FPAR vs leaf area index from a BOREAS site (Chen and Cihlar 1996). The solid line shows the exponential model and the dotted line the linear model used to estimate leaf area index from FPAR. (c) FPAR vs leaf area index from a BOREAS site (data provided by P. Rich, K. F. Huemmrich, and the BOREAS information system). The data marked “N” indicate measurements from needleleaf vegetation (either black spruce or jack pine) and the data marked “A” indicate measurements from an aspen site. The solid line shows the exponential model to estimate leaf area index from FPAR for needleleaf vegetation, the dotted line the linear model for needleleaf vegetation, and the dashed line the exponential model for deciduous broadleaf vegetation. Note that the fit of the linear model is inadequate for both (b) and (c).

  • View in gallery

    The x axis: mean annual NDVI averaged per biome type; the y axis: temporal standard deviation of NDVI anomaly time series aggregated per biome. Numbers indicate the SiB2 land cover types (Table 1).

  • View in gallery

    Interannual variation for 1982–90 in crop yield data of selected sites and in annual global NPP (Gt carbon), based on CASA model runs using different NDVI versions. (a) Time series of residuals from regression between yield and NPP estimates from FASIR, Pathfinder, and GIMMS NDVI for a Turkish agricultural site. (b) Same, but for a Polish site. (c) Global estimates of NPP from FASIR, Pathfinder, and GIMMS data. (d) Optical thickness (τ) estimates at the equator from stratospheric aerosols by the El Chichón eruption and equatorial crossing time (ECT) of NOAA-7, -9, and -11. Analysis of variance estimating global NPP as a function of τ and ECT indicates significance levels of [Pr(τ) = 0.04, Pr(ECT) = 0.07] for global FASIR NPP, of [Pr(τ) = 0.007, Pr(ECT) = 0.005] for global Pathfinder NPP, and of [Pr(τ) = 0.02, Pr(ECT) = 0.0009] for global GIMMS NPP (data from Malmström et al. 1997).

  • View in gallery

    Interannual variation in departures from monthly mean leaf area index (LAI; thick continuous line), precipitation (boxes), and land surface air temperature (thin dotted line) averaged over selected parts of the growing season. Sea surface data are identified with the land cover classification map and were excluded. (a) Western Europe from 64°N, 3°W–46°N, 13°E for early spring (Mar, Apr, May). (b) Same but for summer (Jun–Sep). (c) Nordeste (Brazil) 3°S, 42°W–12°S, 44°W for the entire year. (d) Sahel 17°N, 7°W–12°N, 23°E for Jul–Sep.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 294 251 13
PDF Downloads 144 120 8

A Global 9-yr Biophysical Land Surface Dataset from NOAA AVHRR Data

View More View Less
  • 1 Science Systems and Applications, Inc., and Biospheric Sciences Branch, NASA Goddard Space Flight Center, Greenbelt, Maryland
  • | 2 Biospheric Sciences Branch, NASA Goddard Space Flight Center, Greenbelt, Maryland
  • | 3 Johnson Space Center, Houston, Texas
  • | 4 Department of Botany and Plant Pathology, Michigan State University, East Lansing, Michigan
  • | 5 Department of Geography, University of Maryland, College Park, Maryland
  • | 6 Department of Meteorology, University of Maryland, College Park, and NASA Goddard Space Flight Center, Greenbelt, Maryland
  • | 7 Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado
© Get Permissions
Full access

Abstract

Global, monthly, 1° by 1° biophysical land surface datasets for 1982–90 were derived from data collected by the Advanced Very High Resolution Radiometer (AVHRR) on board the NOAA-7, -9, and -11 satellites. The AVHRR data are adjusted for sensor degradation, volcanic aerosol effects, cloud contamination, short-term atmospheric effects (e.g., water vapor and aerosol effects ⩽2 months), solar zenith angle variations, and missing data. Interannual variation in the data is more realistic as a result. The following biophysical parameters are estimated: fraction of photosynthetically active radiation absorbed by vegetation, vegetation cover fraction, leaf area index, and fraction of green leaves. Biophysical retrieval algorithms are tested and updated with data from intensive remote sensing experiments. The multiyear vegetation datasets are consistent spatially and temporally and are useful for studying spatial, seasonal, and interannual variability in the biosphere related to the hydrological cycle, the energy balance, and biogeochemical cycles. The biophysical data are distributed via the Internet by the Goddard Distributed Active Archive Center as a precursor to the International Satellite Land Surface Climatology Project (ISLSCP) Initiative II. Release of more extensive, higher-resolution datasets (0.25° by 0.25°) over longer time periods (1982–97/98) is planned for ISLSCP Initiative II.

Corresponding author address: Sietse O. Los, Science Systems and Applications, Inc., Biospheric Sciences Branch, Code 923, NASA Goddard Space Flight Center, Greenbelt, MD 20771.

Email: sietse@scirocco.gsfc.nasa.gov

Abstract

Global, monthly, 1° by 1° biophysical land surface datasets for 1982–90 were derived from data collected by the Advanced Very High Resolution Radiometer (AVHRR) on board the NOAA-7, -9, and -11 satellites. The AVHRR data are adjusted for sensor degradation, volcanic aerosol effects, cloud contamination, short-term atmospheric effects (e.g., water vapor and aerosol effects ⩽2 months), solar zenith angle variations, and missing data. Interannual variation in the data is more realistic as a result. The following biophysical parameters are estimated: fraction of photosynthetically active radiation absorbed by vegetation, vegetation cover fraction, leaf area index, and fraction of green leaves. Biophysical retrieval algorithms are tested and updated with data from intensive remote sensing experiments. The multiyear vegetation datasets are consistent spatially and temporally and are useful for studying spatial, seasonal, and interannual variability in the biosphere related to the hydrological cycle, the energy balance, and biogeochemical cycles. The biophysical data are distributed via the Internet by the Goddard Distributed Active Archive Center as a precursor to the International Satellite Land Surface Climatology Project (ISLSCP) Initiative II. Release of more extensive, higher-resolution datasets (0.25° by 0.25°) over longer time periods (1982–97/98) is planned for ISLSCP Initiative II.

Corresponding author address: Sietse O. Los, Science Systems and Applications, Inc., Biospheric Sciences Branch, Code 923, NASA Goddard Space Flight Center, Greenbelt, MD 20771.

Email: sietse@scirocco.gsfc.nasa.gov

1. Introduction

The spatial distribution and temporal dynamics of land surface vegetation are important to understand global interactions among the biosphere, climate, the hydrological cycle, the energy balance, and biogeochemical cycles. Los et al. (1994) and Sellers et al. (1996b) were the first to derive land surface parameters with realistic seasonal and spatial variations for the globe for one year (1987) from normalized difference vegetation index (NDVI) data collected by the Advanced Very High Resolution Radiometer (AVHRR) aboard a National Oceanic and Atmospheric Administration (NOAA) satellite.

Estimation of land surface vegetation parameters from satellite is based on the spectral properties of vegetation; vegetation strongly absorbs visible light, using the energy for photosynthesis, and strongly reflects near-infrared radiation. The NDVI exploits this response: NDVI = (ρvisρnir)/(ρvis + ρnir), where ρvis is the visible land surface reflectance and ρnir is the near-infrared land surface reflectance. NDVI is related most closely to the fraction of photosynthetically active radiation absorbed by the green parts of the vegetation, or FPAR (Sellers 1985; Tucker and Sellers 1986). FPAR is linked closely to the maximum photosynthetic capacity of vegetation.

Temporal and spatial consistency in raw AVHRR NDVI data is seriously compromised by sensor degradation, clouds, viewing geometry, and atmospheric effects (Goward et al. 1991; Los et al. 1994; Gutman and Ignatov 1995). Los et al. (1994) and Sellers et al. (1996b) developed corrections for these effects to improve the spatial and temporal integrity of the 1987 NDVI data.

In this paper, the 1987 record of Sellers et al. (1996b) is expanded to 1982–90. The source data—the 1° by 1° NDVI and 1° by 1° land-cover classification—are discussed in section 2. Corrections to the 1° by 1° NDVI data—collectively referred to as FASIR corrections—are evaluated and updated in section 3 (which see for expansion of FASIR acronym). The updated retrievals of biophysical parameters—FPAR, total leaf area index LT, fraction of green leaves fG, and canopy cover fraction fV—are discussed in section 4. The interannual variation in the data is evaluated in section 5. Sensitivities of photosynthesis and climate parameters to changes in vegetation boundary conditions in a climate model are illustrated in section 6. The merits and limitations of the current datasets are discussed and put in perspective with recent developments in section 7.

The datasets described in this paper are used in a wide variety of applications: in coupled atmosphere–biosphere models (Randall et al. 1996; Sellers et al. 1996ac; Bounoua et al. 2000), in atmospheric tracer models (Denning et al. 1995; Fung et al. 1997), in biogeochemical models (Potter et al. 1993; Field et al. 1995), for validation of biome models (Haxeltine and Prentice 1996), for model intercomparisons (Dirmeyer 1999; Cramer et al. 1999), and to study linkages between global vegetation and climate observations (Los 1998b). The multiyear datasets reflect meaningful interannual variation in the biosphere, and their analysis should increase understanding of, for example, the temporal and spatial distribution of terrestrial carbon sources and sinks (Fung et al. 1987; Keeling et al. 1995), and of soil moisture.

The expanded dataset is distributed by the Goddard Space Flight Center (http://islscp2.gsfc.nasa.gov/) as part of the International Satellite Land Surface Climatology Project (ISLSCP) Initiative II. This prerelease contains monthly NDVI, FASIR-NDVI, FPAR, LT, fV, and fG for 1982–90 at 1° by 1° resolution. A beta version of the ISLSCP land surface dataset for 1982–97 is planned by summer 2000. These datasets contain monthly NDVI, FASIR-NDVI, FPAR, LT, fV, fG, albedo, roughness length, and soil background reflectance (visible and near infrared) for three dominant land cover classes at both 1° by 1° and 0.25° by 0.25° resolution. The beta version will be tested in hydrometeorological and biogeochemical models and will be compared withmoderate resolution imaging spectroradiometer (MODIS) products. A final version will be released by summer 2002.

The purpose of the current document is to provide a complete description of the 9-yr land surface dataset and to provide guidance about its use.

2. Data

The corrected global NDVI data and biophysical parameters were estimated from the following two datasets: 1) the Global Inventory Monitoring and Modeling System (GIMMS) NDVI data (Los et al. 1994) were used as the primary data source, and 2) the land cover classification of DeFries and Townshend (1994) was used to account for biome dependencies in the NDVI corrections (section 3) and in retrieval of biophysical parameters (section 4). The GIMMS NDVI data were available for 1982–90. Aerosols from the Mount Pinatubo eruption compromised the quality of first-generation GIMMS NDVI data between 1991 and 1993. A second-generation GIMMS NDVI dataset is corrected for these effects and will be used to expand the time series to 1998 for a future release.

a. GIMMS NDVI data

The GIMMS monthly continental NDVI data from 1982 to 1990 (Los et al. 1994; Tucker et al. 1994) were used to estimate global biophysical parameters. The continental data, with about 8-km resolution, were averaged spatially over a 1° by 1° grid, excluding surface water (Los et al. 1994). The 1° by 1° resolution meets the demands of most global climate and biogeochemical models. The errors introduced by averaging NDVI prior to estimating biophysical parameters are small (DeFries et al. 1997). Because of limitations in computing resources during the early stages of the GIMMS program, data on component channels, viewing geometry, atmospheric constituents, and cloud frequency were discarded. The GIMMS NDVI data were adjusted for sensor degradation and the effects of volcanic aerosols from the El Chichón eruption in Chiapas, Mexico, during April 1982.

1) Sensor degradation

Sensor degradation in the AVHRR results in gradual changes in the observed reflectance over its lifetime and leads to discontinuities between successive instruments (Fig. 1a) (Price 1987). The GIMMS NDVI data were adjusted for sensor degradation and calibration differences between satellites with an accurate approximation (Los 1998a).

2) Volcanic aerosols

Atmospheric scattering and absorption depend on atmospheric composition, spectral properties of the land surface, viewing angle, and solar zenith angle. Volcanic aerosols affect observation of the land surface over large areas for several years and decrease the NDVI, primarily because they increase atmospheric scattering of solar radiation.

The global NDVI data were corrected for atmospheric aerosols from the El Chichón eruption in April 1982. Because of the high mixing rates and strong zonal component of the wind in the upper atmosphere, we assumed aerosol optical thickness to be dependent on latitude and time and independent of longitude. This assumption is reasonable when several months have passed since the eruption.

A relationship was derived between the optical thickness data over the Pacific Ocean (Vermote et al. 1997a) and deviations in the NDVI values over the land surface for May 1982 to December 1983. The deviation in NDVI was estimated for latitudinal bands from areas with a monthly average NDVI larger than 0.5. In general, areas with these high average NDVI values exhibit low interannual variation (section 5), and it is assumed therefore that the deviations in NDVI data from these areas were caused by the El Chichón aerosols. The threshold value of NDVI greater than 0.5 was selected as a compromise to select green invariant targets and to obtain sufficient spatial and temporal coverage.

Top-of-the-atmosphere reflectance in individual channels increases exponentially with increasing optical thickness. Nonlinear regression (Bates and Watts 1988) was applied to estimate the effect of optical thickness on the NDVI (Fig. 1b) with a similar relationship:
maxβ1eβ2τ
where Δmax is the maximum deviation in NDVI, β1 and β2 are coefficients estimated with nonlinear regression (−0.0874 and −13.44), and τ is aerosol optical thickness. The deviation in NDVI (ΔNDVI) was largest for high NDVI values (ΔNDVI = Δmax) and was negligible over bare soils (ΔNDVI = 0). Based on the results of Vermote et al. (1997a), the deviation in NDVI was assumed to vary linearly with NDVI:
i1525-7541-1-2-183-e2
where NDVImax is equal to 0.54 (the mean NDVI of sites with monthly mean NDVI > 0.5), and NDVImin is equal to 0.0339 (Table 1). NDVI data were adjusted with time- and latitude-varying optical thickness data.

b. Land cover classification

In Sellers et al. (1996b), a global land cover classification based on conventional ground surveys was used for biome-dependent NDVI adjustments and land surface parameter estimation. This classification consisted of 12 land cover types (Table 1). To obtain greater consistency with the NDVI data in the current study, a global land cover classification was derived from the Fourier-adjusted NDVI time series discussed in section 3 (DeFries and Townshend 1994). NDVI time series with classification-dependent corrections (section 3) were not used, to avoid circular reasoning. The land cover classification of DeFries and Townshend was combined with the landcover classification of Collatz et al. (1998) to separate grasslands with different photosynthetic mechanisms (C3 and C4 species). Global land cover classifications from AVHRR data also exist at higher resolutions: for example, at 1-km (Loveland and Belward 1997) and 8-km (DeFries et al. 1998) resolution. Improved classifications based on higher–spatial resolution data, multiple bands, and multiple years are planned for future updates. It is expected that changes caused by land cover classification will be small because the corrections to the NDVI and the relationships between FPAR and leaf area index are first-order dependent on the NDVI and second-order dependent on the classification.

3. Evaluation of FASIR corrections

AVHRR NDVI data are affected by cloud cover (Fig. 1c), atmospheric water vapor (Fig. 1d), variations in viewing angle (Fig. 1e), and solar zenith angle (Fig. 1f). To address interferences in the global NDVI data, a sequence of several corrections was applied, a sequence collectively referred to as FASIR corrections (Sellers et al. 1996b). FASIR is an acronym for Fourier adjustment of outliers in NDVI time series, solar zenith angle correction, interpolation of missing data, and reconstruction of NDVI values over tropical rain forests that have almost continuous cloud cover (Fig. 2). The Fourier adjustment, interpolation, and reconstruction are similar to the techniques discussed in Sellers et al. (1996b), and the solar zenith angle correction was revised.

a. Fourier adjustment

The Fourier adjustment operates on one NDVI time series of a particular grid cell at a time (Figs. 2a,b; Sellers et al. 1996b). It identifies and restores outliers caused by cloud interferences (Fig. 1c) and short-term (⩽2 months) atmospheric effects (Fig. 1d). To correct the full 9-yr time series, the Fourier adjustment is repeated on time series of 12 months, with six months of overlap between successive runs. Only the middle 6 months of each run are used, to avoid leakage effects at the tails of the time series.

The Fourier adjustment was evaluated by removing one month of global NDVI data from a sequence of 12 months. Missing values were estimated with the Fourier adjustment and compared with the original values. This test was applied, leaving out data from January, April, July, and October of 1987, to check seasonally dependent biases.

Close agreement between the estimated and observed NDVI values was found (Fig. 3): the median difference is between −0.004 and −0.002 NDVI, and the spread, expressed as the difference between the 5th and 95th percentiles of the difference distribution, is, at most, 0.1 NDVI units. A slight skewness exists in the difference distributions of April, July, and October; in April the skewness is positive; in July and October the skewness is negative. The standard error of all estimates combined was 0.05 NDVI.

b. Solar zenith angle correction

The solar zenith angle correction adjusts for bidirectional reflectance distribution function (BRDF) effects in the NDVI data that are related to variations in time of overpass of the NOAA satellite. The combined BRDF of land surface and atmosphere generally leads to a decrease in NDVI with increasing viewing angle and solar zenith angle because the atmospheric BRDF is stronger (Figs. 1e,f). Viewing angle effects are reduced in the NDVI data (Holben 1986; Gutman 1991), but this reduction is not the case for solar zenith angle effects.

Solar zenith angle effects are estimated from land cover types with tall vegetation and high NDVI (classes 1 through 6, and 12) and land cover types with bare soil and shrubs and low NDVI (classes 9 and 11; Table 1). Areas with consistently high or low NDVI have low interannual variations, which improves the estimation of solar zenith angle effects. The monthly NDVI data for 1982–90 were stratified into groups of equal vegetation type (section 2) and equal solar zenith angle interval of 2°. The analysis was restricted to areas with noon solar zenith angles smaller than 35° to avoid data from dormant vegetation.

NDVI frequency distributions were calculated for each of the groups with equal vegetation type and solar zenith angle interval. For the tall vegetation classes, the 98th percentiles (representing fully green conditions) were plotted versus the midpoint of the solar zenith angle interval (Fig. 2c). A model proposed by Singh (1988) was fitted with nonlinear regression (Bates and Watts 1988):
98,θ98,0k1θ(π/6)]k2
where NDVI98,0 represents the 98th percentile of the NDVI distribution for a particular class for an overhead sun, θ is the solar zenith angle, and k1 and k2 are coefficients estimated with nonlinear regression (Table 1).

The analysis was repeated for NDVI populations from bare soils and deserts [Simple Biosphere Model, version 2 (SiB2) classes 9 and 11] using the 2d instead of the 98th percentiles. No significant solar zenith angle effect on the NDVI of bare areas was found (see also Los 1998a). This conclusion derived from 9 yr of data is at odds with an earlier analysis (Sellers et al. 1996b) on 1 yr of data, for which a small solar zenith angle effect was found over deserts.

NDVI values were adjusted to values corresponding to a solar zenith angle of 0° using
i1525-7541-1-2-183-e4
The effect of solar zenith angle between 0° and 30° is small (when the hot-spot effect is ignored). The solar zenith angle correction was held constant for solar zenith angles larger than 60° (Fig. 2c) because its estimation was deemed to be unreliable for these angles.

The solar zenith angle effects estimated from the global data were compared with solar zenith angle effects obtained by simulating atmospheric effects over land surface measurements. The land surface BRDF was obtained by fitting the Roujean model (Roujean et al. 1992) through NDVI data measured from multiple viewing and solar zenith angles at the First ISLSCP Field Experiment (FIFE) site (Deering and Leone 1986). The atmospheric BRDF effects were simulated with The Second Simulation of the Satellite Signal in the Solar Spectrum (6S) code (Vermote et al. 1997b) for a continental midlatitude atmosphere with optical thicknesses of 0.5 and 0.1 (Fig. 1f). The statistical analysis of global NDVI data shows a dependency of NDVI on solar zenith angle for SiB2 classes 1, 6, 7, 8, 9, 11, and 12 between the two scenarios for different atmospheres simulated with 6S (Fig. 1f), indicating that the top-of-the-atmosphere solar zenith angle effect is dominated by the radiative properties of the atmosphere. The solar zenith angle effect estimated for classes 2, 3, 4, 5, and 10 is slightly stronger. These estimates of solar zenith angle effect are in between those reported by Privette et al. (1995) for a clear and hazy sky and are smaller than those reported by Singh (1988).

c. Miscellaneous effects

The NDVI of needleleaf evergreen land cover types during winter was set by extrapolating the October NDVI value (Fig. 2d). This method assumes that, in October, the satellite-measured greenness is predominantly from evergreen needle trees since deciduous trees have shed their leaves. This interpolation is necessary to meet model requirements of spatially and temporally continuous data and to avoid high albedo values in boreal forests that lead to unrealistic predictions of surface temperature (Betts et al. 1998). Overestimation of the NDVI for these higher latitudes outside the growing season is unlikely to lead to errors in evapotranspiration or assimilation rates in models because the vegetation is physiologically inactive (Los 1998b).

The reconstruction of NDVI time series for the tropical evergreen broadleaf vegetation is necessary to adjust for the effects of persistent cloud cover and atmospheric water vapor (Fig. 2e). Low NDVI values result in low FPAR and leaf area index estimates, which introduce errors into the calculation of photosynthesis, evapotranspiration, interception, and the energy balance (Los 1998b). A side effect of the procedure is that all seasonality in the data is eliminated and areas incorrectly classified as evergreen broadleaf will have high NDVI values throughout the year. The procedure does provide an overall improvement in that it diminishes the number of outliers, especially in areas with continuous clouds.

The following effects were not accounted for explicitly in the FASIR-NDVI data. 1) The average registration error of 6 km (Baldwin and Emery 1993; Holben 1986) is negligible at 1° by 1° resolution. 2) Topographic effects are reduced greatly in low-resolution data (Burgess et al. 1995). 3) Variations in the NDVI of the Sahara related to variations in soil background reflectance have a standard deviation of about 0.02 NDVI (Sellers et al. 1996b). These spatial variations are less important for the temporal variation of the NDVI. 4) Across-track lines of anomalous values in AVHRR imagery were removed by visual inspection of the data. These anomalous values may be caused by problems at the ground receiving station, such as interferences of nearby buildings and trees.

4. Evaluation of FPAR and leaf area index models

In Sellers et al. (1996b), biophysical parameters were estimated from NDVI. NDVI was converted into simple ratio (SR); FPAR was estimated from SR with a linear function; and leaf area index was estimated from FPAR with a logarithmic function, a linear function, or a combination. These relationships of Sellers et al. (1996b) are evaluated and updated based on observations from FIFE, the Oregon Transect Ecosystem Research project (OTTER), the Boreal Ecosystem–Atmosphere Study (BOREAS), and the Hydrological Atmospheric Pilot Experiment (HAPEX)—Sahel. The estimation of vegetation cover fraction fV was introduced to improve relationships between FPAR and LT (Fig. 4).

a. FPAR

In Sellers et al. (1996b), relationships between FPAR and NDVI were derived by land cover class. Maximum and minimum NDVI were related to maximum and minimum FPAR according to
i1525-7541-1-2-183-e5
where SR = (1 + NDVI)/(1 − NDVI), SRrange = SRmax − SRmin corresponds to the difference between the 98th and 2d percentiles of the SR frequency distributions (Table 1), and FPARrange = FPARmax − FPARmin = 0.95 − 0.01. Equation (5) is referred to as the SR–FPAR model. An alternative model, referred to as the NDVI–FPAR model (Choudhury 1987; Goward and Huemmrich 1992) is given by
i1525-7541-1-2-183-e6
where NDVIrange = NDVImax − NDVImin corresponds to the difference between the 98th and 2d percentiles of the NDVI frequency distributions (Table 1).

Both models were tested on data from FIFE (Hall et al. 1992), OTTER (Angelici et al. 1991), BOREAS (Fournier et al. 1997; Chen et al. 1997; Chen 1996), and HAPEX—Sahel (Hanan et al. 1997). The 98th percentiles of NDVI distributions at the site were calculated and linked to a maximum FPAR of 0.95. For FIFE, where FPAR values were lower than 0.95, the NDVI corresponding to an FPAR of 0.95 was estimated from the scattering-by-arbitrarily-inclined-leaves (SAIL) model (Fig. 4 of Sellers et al. 1996b). For HAPEX (maximum FPAR below 0.5), the maximum NDVI values from the FIFE site were used, analogous to the global analysis in which short vegetation types were lumped together. The minimum FPAR value of 0.01 was related to the bare soil NDVI in the global dataset, because bare soil NDVI values were not measured at the ground. FPAR values were calculated with the NDVI–FPAR and the SR–FPAR models for all sites and were combined (Figs. 4b, 5).

Model performance was tested with an analysis of variance (Draper and Smith 1981). The data were split into eight groups of equal size, and, for each group, the deviations from the mean were compared with the deviations from the 1:1 line. The NDVI–FPAR model performed worst (rms = 0.19) and resulted in significant overestimates of FPAR (Fig. 5a). The NDVI–SR model performed better (rms = 0.11) but significantly underestimated FPAR (Fig. 5b). A third, intermediate model, calculating the average FPAR of the NDVI–FPAR and SR–FPAR models, performed better (rms = 0.08) and did not show a lack of fit at the 1% significance level (Table 2).

In this analysis of FPAR–NDVI relationships, the effect of the atmosphere was ignored. Figure 6 shows the relationship between uncorrected NDVI and the NDVI corrected for Rayleigh scattering and ozone absorption in the atmosphere. The atmospherically corrected NDVI differs from the uncorrected NDVI by a multiplication factor. The root-mean-square error of this simple model is about 0.01. This error is one order of magnitude smaller than the error in the SR–FPAR model. Short-term variations in water vapor and aerosols are in part accounted for by the Fourier adjustment. Water vapor and aerosol climate descriptions are not addressed explicitly but are, in part, accounted for by the statistical selection of NDVI values that correspond to maximum vegetation development. Their contribution to the overall error in the FPAR estimates is expected to be small.

b. Leaf area index

The relationships between total leaf area index LT and FPAR were evaluated with data from the FIFE and BOREAS experiments. In Sellers et al. (1996b), a linear relationship was used between FPAR and green leaf area index LG for vegetation types 4, 5, and 9, and a logarithmic equation was used for other vegetation types. The FIFE data show close agreement between the modeled and the measured FPAR versus total (live and dead) leaf area index (Fig. 7b).

For needleleaf biomes, a linear relationship betweenFPAR and LG was derived from a spatial analysis (Sellers et al. 1996b). The BOREAS data indicate that this linear relationship is invalid at the canopy level (Figs. 7b,c). To vary the leaf area index as a linear function of FPAR spatially (Sellers et al. 1992) and as a logarithmic function of FPAR at the canopy level, the vegetation cover fraction is varied by grid cell according to (Fig. 4c)
fVt
where fV is the vegetation cover fraction, and max(FPARt) is the maximum FPAR over the year. FPAR values of the total grid cell are converted to FPAR estimates for the cover fraction:
FPARfVfV
Green LG for the vegetation cover fraction LG,fV is expressed as (Fig. 4d)
i1525-7541-1-2-183-e9
here i indicates that maximum LG depends on vegetation type. The green leaf area index for the total grid cell is
LGLG,fVfV
Two different cases are distinguished for the estimation of dead leaf area index. When leaf area index increases, the dead leaf area index LD for the total grid cell is estimated with
LDLS
where LS is the stem area index. When leaf area index decreases, the dead leaf area index for the grid cell is estimated with
LDLG,fV,t−1LG,fV,tfV
The total leaf area index for the grid cell is then
LTLG,fVfVLD
The greenness fraction of green leaf area index to the total leaf area index fG, which equals LG/LT, is independent of fV.

5. Interannual variation

Interannual variations in the biosphere could provide important clues to relationships among the biosphere, the carbon cycle, the water cycle, and the energy balance. The realism of interannual variations in the 9 yr of NDVI data is evaluated in the current section.

a. Interannual variation by land cover type

The magnitude of interannual variation varies strongly among land cover types. Figure 8 compares the mean annual NDVI and the interannual variation, expressed as the standard deviation of the departures from the monthly mean NDVI, for each of the 12 land cover types. Deserts (class 9) and tropical, evergreen, broad leaf cover types (class 1) both exhibit small interannual (and seasonal) variation, with deserts having low NDVI values and tropical forests high NDVI values throughout the year (Fig. 8). The largest interannual range is found in high latitude biomes (classes 3, 5, and 10; mixed trees, needle leaf deciduous, and tundra, respectively) indicative of a short, highly variable growing season. Semiarid, seasonal grasslands (class 7) have an interannual range similar to that of open woodland (class 6) but have lower annual means.

b. Interannual variation in net primary production fields

An effective approach for evaluating interannual variation in NDVI time series is to use a carbon model to convert the NDVI into a time series of net primary production (NPP) estimates. The NPP estimates then can be compared with appropriate ground-based production data. Malmström et al. (1997) evaluated NPP estimates produced by the Carnegie–Ames–Stanford Approach (CASA) model (Potter et al. 1993; Field et al. 1995) from several different NDVI time series: Le Centre d’Etudes Spatiales de la Biosphère (CESBIO) (Ruimy et al. 1994), Pathfinder (James and Kalluri 1994), FASIR (this study), and a modified GIMMS NDVI (Los et al. 1994) [referred to as “pre” FASIR in Malmström et al. (1997)]. NPP estimates based on the different NDVI time series were tested against yield data from seven agricultural regions and one boreal forest tree ring index. The agricultural sites were selected on the basis of size (11–58 1° by 1° grid cells), crop homogeneity, absence of irrigation, and data availability. The tree ring site (45 plots for about 20 000 ha) was selected on the basis of the availability of stand-level tree ring data from the two main physiognomic tree groups and on the basis of detailed land cover maps.

The global NPP estimates show substantially different interannual variability for 1982–90 (Fig. 9). A significant amount of variance in the GIMMS and Pathfinder estimates of global NPP is explained by the equatorial crossing time of the NOAA satellites—and associated increased solar zenith angles—and by the increase in optical thickness caused by stratospheric aerosols from the El Chichón eruption. The variance explained by these effects was reduced significantly in the FASIR NPP. Solar zenith effect was strongest in the GIMMS NPP, slightly less in the Pathfinder NPP, and not significant at the 5% level in the FASIR NPP. The ordering agrees with the corrections applied: the GIMMS data are not corrected for solar zenith angle effects, the ozone and Rayleigh corrections applied to the Pathfinder data partly correct for solar zenith angle effects, and the FASIR data have an explicit correction for the combined BRDF effects of atmosphere and land surface (Table 3). The optical thickness data explained a significant amount of variance in the Pathfinder data. The amount of variance explained by optical thickness in the FASIR and GIMMS data was much reduced, below the 1% level in both cases. The amount of variance explained by optical thickness in the GIMMS data was slightly higher than in the FASIR NPP, possibly because the Fourier adjustment reduced residual aerosol effects.

The tests against ground-based data further indicate the importance of solar zenith angle corrections and volcanic aerosol corrections. In the agricultural tests, the FASIR- and GIMMS-based NPP estimates were correlated strongly with yield at five of the seven sites and somewhat correlated at another. The Pathfinder-based estimates were correlated strongly with yield at only three sites and somewhat correlated at two. At the boreal forest site, the FASIR, GIMMS, and Pathfinder estimates were significantly correlated with the interannual variability in the tree ring index, but only if the NPP estimates were detrended (see below). For solar zenith angle, the tests showed that the FASIR processing improved the fit between estimates and yield data by eliminating some deviations corresponding to changes in overpass time. At the Turkish and Polish sites, for example, the NPP estimates based on NDVI versions without explicit solar zenith angle corrections (GIMMS, Pathfinder) deviated from the yield data in parallel with NOAA-9’s drift to later crossing times in 1986–88 (Fig. 9); in contrast, the FASIR estimates closely reproduced the yield data. Note that a pattern of systematic mismatches between NPP estimates and yield data for 1983–84 for all the NDVI versions suggests an unresolved problem with data from NOAA-7 (Fig. 9).

During 1982–90, there were positive annual trends in the NPP estimates for most vegetation classes. The trends were smaller in the FASIR-based estimates than in the estimates based on the other NDVI versions, except for the boreal regions, where the Pathfinder trends were smaller (Table 4). Both the trend in the Polish yield data and the absence of trend at the other agricultural sites were reproduced in the FASIR estimates. However, there was a +4.7% yr−1 to +5.0% yr−1 trend evident in all of the NPP estimates for the boreal forest site that was not evident in the tree ring data. Additional investigation of production patterns in forested regions is warranted.

Comparison of the global trends in the FASIR-based NPP estimates with independent estimates based on atmospheric data indicates that these NPP trends are two to four times higher than is implied by the atmospheric carbon dioxide record (Randerson et al. 1996; Malmström et al. 1997).

c. Interannual variation in leaf area index

In this section the interannual variation in leaf area index is evaluated for selected areas. The interannual variations in the FASIR–NDVI and FPAR data are similar to those in the leaf area index. Independent, in situ measurements to assess the realism of interannual variation in the leaf area index data do not exist. Therefore, the leaf area index anomalies are compared with anomalies in precipitation (Eischeid et al. 1991) and land surface air temperature (Jones et al. 1985). Since temperature and precipitation are critical to vegetation growth, their interannual variations provide important information about the realism of interannual variation in the LAI data. The sign and magnitude of these associations are region dependent.

Leaf area index, land surface air temperature, and precipitation anomalies were analyzed for three selected regions (Fig. 10). A principal component analysis was applied to find common factors in the precipitation, temperature, and leaf area index data in Fig. 10. The loadings on the first principal component, that is, a measure for shared variance among parameters, and the amount of total variance explained by the first principal component are shown in Table 5.

In the first example (Fig. 10a; Table 5, first column) mean leaf area index, precipitation, and temperature anomalies are analyzed for early spring in western Europe. The interannual variations in leaf area index and temperature anomalies are similar: below- (above-) normal temperatures occur with below- (above-) normal leaf area index values. Precipitation has a weaker, negative association with both temperature and leaf area index (Table 5). Thus warmer springs tend to have more vegetation activity than do colder ones. Moisture is likely to be sufficient during spring, hence variations in precipitation do not affect vegetation directly. An indirect effect of precipitation is a cooling of the land surface, and this cooling can inhibit vegetation growth. The data, furthermore, show an increase during the 1980s for both leaf area index and temperature.

The summer over western Europe shows a different association among temperature, rainfall, and leaf area index anomalies (Fig. 10b). Leaf area index and temperature anomalies are negatively correlated and rainfall is positively correlated with leaf area index anomalies (Table 5). A decrease in rainfall results in reduced moisture for plants and reduced evapotranspiration, which causes the temperature to increase. Notice that, in contrast to Fig. 10a, neither temperature nor leaf area index anomaly shows an increase during the summer, hence increased temperatures during spring may lead to a longer growing season but not necessarily to increased vegetation activity throughout the year.

Variations in rainfall in Nordeste (Brazil; Figure 10b) are associated with the occurrences of warming and cooling of the western equatorial Pacific Ocean generally referred to as warm and cold El Niño–Southern Oscillation, or ENSO, events (Ropelewski and Halpert 1987). Warm ENSO events are associated with decreased convection and droughts in Nordeste, and cold ENSO events are associated with periods of increased convection and precipitation. Reduced rainfall, which occurred during the warm ENSO events of 1982/83 and 1987/88, leads to reduced soil moisture and vegetation activity (Myneni et al. 1995), and this reduction leads to reduced evapotranspiration and increased air surface temperatures. This negative association between air temperature anomalies on the one hand and leaf area index and rainfall anomalies on the other is shown in Fig. 10c and Table 5.

During 1984, a large drought occurred in the Sahel, the area south of the Sahara (Fig. 10d). After 1984, the rains returned and the vegetation recovered (Tucker et al. 1994). During the 1984 drought, leaf area index decreased and temperature increased, in part because of reduced plant activity and evapotranspiration (Table 5).

The interannual variation in leaf area index for selected sites is consistent with the interannual variations found in rainfall and temperature. This consistency gives further support for the realism of interannual signals in the vegetation data.

6. Sensitivity of hydrometeorological parameters

The sensitivity of elements of the carbon cycle, water cycle, and energy balance to changes in vegetation was tested with the SiB2 Colorado State University (CSU) GCM (Sellers et al. 1996a; Randall et al. 1996). The sensitivity to two different land surface scenarios was evaluated: scenarios with minimum and maximum fraction, respectively, of photosynthetically active radiation absorbed by the green part of the vegetation (FPAR). The minimum and maximum FPAR scenarios were derived from the 1982–90 global FPAR dataset.

The sensitivity test with the minimum and maximum FPAR runs has two objectives. The first is to establish which parameters are most sensitive to changes in vegetation and thus require the most accurate vegetation boundary conditions; the second is to explore the mechanisms behind the sensitivity of water, carbon, and energy fluxes to changes in vegetation.

a. Design of the experiment

The boundary conditions for the minimum and maximum FPAR scenarios were created by selecting, for each month and grid cell, the monthly minimum and maximum FPAR respectively (Bounoua et al. 2000). The 12 monthly maximum and minimum FPAR fields represent extremes of observed conditions over the 9-yr period. The range between these extremes is about 10 times larger than the range between the years with global maximum and minimum FPAR, because local extremes were not concurrent for all areas.

The SiB2 GCM was run for 30 yr to obtain initial sea surface temperature and soil moisture conditions. The minimum and maximum FPAR scenarios then were run over 15 yr from these initial conditions. Averages over the entire year were calculated over the last 5 yr of the runs. A similar design was used for a more in-depth climate sensitivity study (Bounoua et al. 2000).

b. Sensitivity analysis

Table 6 summarizes the results averaged over the land surface in the Tropics (30°S–30°N), in mid-to-high northern latitudes (40°–70°N), and for the globe. Changes in sign as a result of increased FPAR are consistent for all parameters with the exception of albedo. Albedo slightly increased in the Tropics and decreased in higher latitudes. The increase in albedo in the Tropics is the result of prescribed soil background reflectance values that are slightly lower than is the canopy reflectance. The decrease in albedo with FPAR in mid-to-high latitudes is largely the result of variations in snow cover.

An increase in FPAR results in an increase in carbon assimilation (Table 6). Increased carbon assimilation (photosynthesis) is related directly to increased transpiration by vegetation. Evaporation also increases because leaf area index increases and more precipitation is intercepted by leaves. The increase in total evapotranspiration results in a depletion of soil water, a decrease in surface runoff, and an increase in atmospheric water vapor. Increased atmospheric water vapor causes an increase in land surface precipitation and provides a small negative feedback to the initial depletion of soil moisture. The increase in (mostly convective) precipitation is much stronger in higher latitudes than it is in the Tropics.

Although variations in albedo with FPAR are small, there are changes in the land surface energy balance that are associated with increased photosynthesis. Both canopy and air temperature decrease as a result of evaporative cooling. The main effect of an FPAR increase on the energy balance is a change in the partitioning of the energy balance. The increased FPAR (and increased leaf area index) increases interception of shortwave radiation by the canopy, leading to increased sensible and latent heat fluxes. The increase in leaf area index and associated increase in sensible canopy heat flux more than offset the decrease in temperature as a result of evaporative cooling. The soil is shaded by the plants, and throughfall of precipitation is reduced, hence both sensible and latent heat from the soil are reduced.

The sensitivity analysis indicates that, within the range of observed variations in FPAR, physiological effects on the energy balance are larger than are albedo effects. The land surface hydrological behavior is affected more strongly by variations in vegetation than is the net land surface energy balance. Carbon assimilation also is affected by FPAR variations. Although the minimum and maximum FPAR scenarios represent extreme conditions for the globe, locally these conditions did occur over 1982–90. The changes in parameters summarized in Table 6 probably are realistic at the local scale or as global extremes over longer time periods.

7. Discussion

The following important modifications were made to the the previous study by Sellers et al. (1996b): the 1987–88 record was expanded to 1982–90, the FASIR corrections to improve the spatial and temporal consistency of the NDVI data were updated, algorithms to estimate biophysical parameters were tested and updated, and the sensitivity of several near-surface climatic parameters to changes in FPAR was tested.

a. Biophysical parameters

In comparison with the ISLSCP Initiative I data, FPAR values are higher for all land cover types, and leaf area index values are higher for most land cover types except the clustered ones (classes 3, 4, 5, and 8;Figs. 4, 5). The canopy cover fraction is estimated from the maximum FPAR value during the year. The current formulation of canopy cover fraction brings into agreement observations at the plot level, where leaf area index varies linearly with FPAR (Sellers et al. 1992) and at the canopy level, where leaf area index varies exponentially with FPAR.

The key parameters for most biophysical models, FPAR and leaf area index, are first-order dependent on the FASIR–NDVI data and second-order dependent on the classification (see also DeFries et al. 1997). For example, the NDVI corresponding to an FPAR of 0.95 varies between 0.618 and 0.687 (Table 1), hence land cover type accounts for at most 0%–10% of the variation in FPAR; the maximum variation in leaf area index with land cover type is larger, between 6% and 35%, because the variation in maximum leaf area index values with land cover type is larger. Land cover–dependent effects on interannual variation in biophysical parameters are much smaller.

The authors’ earlier work has led to similar efforts to estimate biophysical parameters from AVHRR data with both simple models (Nemani et al. 1996) and more sophisticated radiative transfer models (Myneni et al. 1997a). Given the range of uncertainty in measurements and in scaling of validation data from ground to pixel level (about 0.08 FPAR), it is doubtful whether more advanced models would give better results than simple models do. Spatial distributions and seasonality of FPAR and leaf area index resemble spatial distributions and seasonality of NDVI in most cases, possibly with the exception of the African equatorial forests, for which estimates with the more advanced model are low (leaf area index between 1.5 and 2.5). The higher spatial resolution in several of these datasets is an advantage for regional models. The datasets from Nemani et al. (1996) and Myneni et al. (1997a) lack corrections for volcanic aerosols and solar zenith angle variations; interannual variation in these datasets therefore is less reliable. Indeed, these datasets are used most suitably as climatological descriptions.

b. Interannual variation

Sensor degradation, cloud effects, and solar zenith angle effects complicate the analysis of interannual variation in NDVI data (Goward et al. 1991; Los et al. 1994; Gutman and Ignatov 1995). These extraneous sources are reduced greatly in the FASIR–NDVI data. Comparison with interannual variations in crop yield and in climate data showed that the FASIR corrections improve the interannual signals in the NDVI data. There are residual errors but they are smaller than the interannual signals. The analysis of the agricultural sites shows that, on average, 40% of the variance in interannual variation of crop yield is not accounted for. This error must be divided among contributions from NDVI, rainfall, temperature, and poor quantification of below-ground NPP in the model. An error of 20%–30% in interannual variation in the NDVI is therefore a reasonable estimate.

Because of the limited length of the time series, we caution about its use for analysis of interannual variation on larger than decadal timescales. We have more confidence in the interannual variation of 2–4-yr cycles (magnitude of about 10%–15% of the seasonal amplitude) than in recently reported trends in northern latitudes with a magnitude of 1%–3% of the seasonal amplitude (Myneni et al. 1997b), which is similar in magnitude to our estimated margin of error. There is supporting evidence for a trend in the data for some areas, for example, the increase in leaf area index during spring in western Europe correlates with land surface air temperature, and, where NPP increased over selected agricultural sites, increased crop yields were found. Data from other areas, for example, from the boreal forest, do not support the trend in the NDVI. Atmospheric constraints suggest that the overall global trend should be smaller (Randerson et al. 1996; Malmström et al. 1997). Better estimates of interannual variation in NDVI data must be obtained from improved datasets over longer time periods. Lacking comparable ground measurements, interannual variations in FASIR–NDVI can be checked for consistency with interannual variations in precipitation and temperature.

c. Future updates

Updated land surface datasets are planned as part of the ISLSCP Initiative II. These datasets will be available for a longer time period (1982–98), at both 1° by 1° spacing and 0.25° by 0.25° spacing. Land surface parameters will be retained for two to four dominant classes in a grid cell to retain some of the information that is lost by resampling the data from 8 km to 0.25° or 1°. Information on fractional cover within a grid cell is important for a correct description of land surface processes in biogeochemical (Fung et al. 1997) and hydrometeorological models. The longer time series in the updated datasets will improve detection of interannual variation in vegetation and help to resolve uncertainties in trend analysis of vegetation. Cross validation with in situ data and MODIS products will allow better integration of the AVHRR and MODIS record. Valuable historic information thus can be preserved and expanded over longer periods.

Acknowledgments

This research was funded by a NASA Earth Observing System-Interdisciplinary Science Grant (Biosphere–Atmosphere Interactions Project), Contracts NAS-531732 and NAS5-99085. The following people aided in data collection or processing:R. Rank, J. Rosenfelder, D. Rosenfelder, and W. W. Newcomb (GIMMS NDVI data); Dr. E. F. Vermote (aerosol optical thickness data); Dr. D. W. Deering (FIFE surface reflectance data); E. T. Kanemasu, B. Blad, A. Nelson, J. Killeen, L. Ballou, T. Shah, and C. Hays (FIFE biophysical properties data); B. L. Blad, E. A. Walter-Shea, C. J. Hays, and M. A. Mesarch (FIFE Surface Reflectance Modular Multiband Radiometer data);Drs. P. Rich and R. Fournier (BOREAS FPAR data); Dr. C. Walthall and the BOREAS helicopter crew (BOREAS reflectance data); and J. M. Chen and J. Cihlar (BOREAS FPAR, leaf area index, and NDVI data). Dr. K. F. Huemmrich gave useful comments on the FIFE, OTTER, and BOREAS data; Dr. G. Gutman and two anonymous reviewers made useful suggestions for improvement of this paper. We gratefully acknowledge the contributions of these many individuals. We thank Dr. G. Asrar of NASA Headquarters for funding this research and for encouraging our work.

REFERENCES

  • Angelici, G. L., J. W. Skiles, and L. Z. Popovici, 1991: OTTER pilot land data system, vol. 1, version 1: Satellite, aircraft and ground measurements. CD-ROM, NASA. [Available from Oak Ridge National Laboratory DAAC, P.O. Box 2008, MS 6407, Oak Ridge, TN 37831.].

  • Baldwin, D. G., and W. J. Emery, 1993: Systematized approach to AVHRR navigation. Ann. Glaciol.,17, 414–420.

    • Crossref
    • Export Citation
  • Bates, D. M., and D. G. Watts, 1988: Nonlinear Regression Analysis and its Applications. John Wiley and Sons, 384 pp.

    • Crossref
    • Export Citation
  • Betts, A. K., P. Viterbo, A. Beljaars, H.-L. Pan, S.-Y. Hong, M. Goulden, and S. Wofsy, 1998: Evaluation of land-surface interaction in ECMWF and NCEP/NCAR reanalysis models over grassland (FIFE) and boreal forest (BOREAS). J. Geophys. Res.,103, 23 079–23 085.

    • Crossref
    • Export Citation
  • Bounoua, L., G. J. Collatz, S. O. Los, P. J. Sellers, D. A. Dazlich, C. J. Tucker, and D. A. Randall, 2000: Sensitivity of climate to changes in NDVI. J. Climate, in press.

    • Crossref
    • Export Citation
  • Burgess, A. W., P. Lewis, and J. P. A. L. Muller, 1995: Topographic effects in AVHRR NDVI data. Remote Sens. Environ.,54, 223–232.

    • Crossref
    • Export Citation
  • Chen, J. M., 1996: Canopy architecture and remote sensing of the fraction of photosynthetically active radiation absorbed by boreal conifer forests. IEEE Trans. Geosci. Remote Sens.,34, 1353–1368.

    • Crossref
    • Export Citation
  • ——, and J. Cihlar, 1996: Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sens. Environ.,55, 153–162.

    • Crossref
    • Export Citation
  • ——, P. M. Rich, S. T. Gower, J. M. Norman, and S. Plummer, 1997:Leaf area index of boreal forests: Theory, techniques, and measurements. J. Geophys. Res.,102, 29 429–29 444.

  • Choudhury, B. J., 1987: Relationships between vegetation indices, radiation, absorption and net photosynthesis evaluated by a sensitivity analysis. Remote Sens. Environ.,22, 209–233.

    • Crossref
    • Export Citation
  • Collatz, G. J., J. A. Berry, and J. S. Clark, 1998: The effect of climate and atmospheric CO2 partial pressure on the global distribution of C4 grasses: Present, past, and future. Oecologia,114, 441–454.

    • Crossref
    • Export Citation
  • Cramer, W., D. W. Kicklighter, A. Bondeau, B. Moore, C. Churkina, B. Nemry, A. Ruimy, and A. Schloss, 1999: Comparing global models of terrestrial net primary productivity (NPP): Overview and key results. Global Change Biol.,5, 1–15.

    • Crossref
    • Export Citation
  • Deering, D. W., and P. Leone, 1986: A sphere-scanning radiometer for rapid directional measurements of sky and ground radiance. Remote Sens. Environ.,19, 1–24.

    • Crossref
    • Export Citation
  • DeFries, R. S., and J. R. G. Townshend, 1994: NDVI-derived landcover classifications at global scale. Int. J. Remote Sens.,15, 3567–3586.

    • Crossref
    • Export Citation
  • ——, ——, and S. O. Los, 1997: Scaling landcover heterogeneity for global atmosphere–biosphere models. Scale in Remote Sensing and GIS, D. A. Quattrochi and M. F. Goodchild, Eds., Lewis Publishers, 231–246.

  • ——, M. Hansen, J. R. G. Townshend, and R. Sohlberg, 1998: Global landcover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiers. Int. J. Remote Sens.,19, 3141–3168.

  • Denning, A. S., I. Y. Fung, and D. A. Randall, 1995: Latitudinal gradient of atmospheric CO2 due to seasonal exchange with land biota. Nature,376, 240–243.

    • Crossref
    • Export Citation
  • Dirmeyer, P. A., A. J. Dolman, and N. Sato, 1999: The Global Soil Wetness Project: A pilot project for global land surface modeling and validation. Bull. Amer. Meteor. Soc.,80, 851–878.

    • Crossref
    • Export Citation
  • Draper, N. R., and H. Smith, 1981: Applied Regression Analysis. John Wiley and Sons, 709 pp.

  • Eischeid, J. K., H. F. Diaz, R. S. Bradley, and P. D. Jones, Eds., 1991:A comprehensive precipitation dataset for global land areas. U.S. Dept. of Energy Carbon Dioxide Research Program Rep. TR051, 82 pp. [Available from Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, TN 37831.].

  • Field, C. B., J. T. Randerson, and C. M. Malmström, 1995: Global net primary production: Combining ecology and remote sensing. Remote Sens. Environ.,51, 74–88.

    • Crossref
    • Export Citation
  • Fournier, R. A., P. M. Rich, and R. Landry, 1997: Hierarchical characterization of canopy architecture for boreal forest. J. Geophys. Res.,102, 29 445–29 454.

    • Crossref
    • Export Citation
  • Fung, I. Y., C. J. Tucker, and K. C. Prentice, 1987: Application of Advanced Very High Resolution Radiometer vegetation index to study atmosphere–biosphere exchange of CO2. J. Geophys. Res.,92, 2999–3015.

    • Crossref
    • Export Citation
  • ——, and Coauthors, 1997: Carbon 13 exchanges between the atmosphere and biosphere. Global Biochem. Cycles,11, 507–533.

    • Crossref
    • Export Citation
  • Goward, S. N., and K. F. Huemmrich, 1992: Vegetation canopy PAR absorbtance and the normalized difference vegetation index: An assessment using the SAIL model. Remote Sens. Environ.,39, 119–140.

    • Crossref
    • Export Citation
  • ——, B. Markham, D. G. Dye, W. Dulaney, and J. Yang, 1991: Normalized difference vegetation index measurements from the Advanced Very High Resolution Radiometer. Remote Sens. Environ.,35, 257–277.

    • Crossref
    • Export Citation
  • Gutman, G. G., 1991: Vegetation indices from AVHRR: An update and future prospects. Remote Sens. Environ.,35, 121–136.

    • Crossref
    • Export Citation
  • ——, and A. Ignatov, 1995: Global land monitoring from AVHRR: Potential and limitations. Int. J. Remote Sens.,16, 2301–2309.

    • Crossref
    • Export Citation
  • Hall, F. G., K. F. Huemmrich, S. J. Goetz, P. J. Sellers, and J. E. Nickeson, 1992: Satellite remote sensing of surface energy balance: Success, failures, and unresolved issues in FIFE. J. Geophys. Res.,97, 19 061–19 089.

    • Crossref
    • Export Citation
  • Hanan, N. P., A. Bégué, and S. D. Prince, 1997: Errors in remote sensing of intercepted photosynthetically active radiation: An example from HAPEX—Sahel. J. Hydrol.,188–189, 676–696.

    • Crossref
    • Export Citation
  • Haxeltine, A., and I. C. Prentice, 1996: BIOME3: An equilibrium terrestrial biosphere model based on ecophysiological constraints, resource availability and competition among plant functional types. Global Biogeochem. Cycles,10, 693–709.

  • Holben, B. N., 1986: Characteristics of maximum-value composite images for temporal AVHRR data. Int. J. Remote Sens.,7, 1435–1445.

    • Crossref
    • Export Citation
  • James, M. E., and S. N. V. Kalluri, 1994: The Pathfinder AVHRR land dataset: An improved coarse resolution dataset for terrestrial monitoring. Int. J. Remote Sens.,15, 3347–3363.

    • Crossref
    • Export Citation
  • Jones, P. D., S. C. B. Raper, B. Santer, B. S. G. Cherry, C. Goodess, P. M. Kelly, T. M. L. Wigley, R. S. Bradley, and H. F. Diaz, Eds., 1985: A grid point surface air temperature dataset for the Northern Hemisphere. U.S. Dept. of Energy Carbon Dioxide Research Division Tech. Rep. TR022, 251 pp. [Available from Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, TN 37831.].

  • Justice, C. O., T. F. Eck, D. Tanré, and B. N. Holben, 1991: The effect of water vapor on the normalized difference vegetation index for the Sahelian region from NOAA-AVHRR data. Int. J. Remote Sens.,12, 1165–1187.

    • Crossref
    • Export Citation
  • Keeling, C. D., T. P. Whorf, M. Wahlen, and J. van der Plicht, 1995:Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980. Nature,375, 666–670.

    • Crossref
    • Export Citation
  • Los, S. O., 1998a: Estimation of the ratio of sensor degradation between NOAA-AVHRR channels 1 and 2 from monthly NDVI composites. IEEE Trans. Geosci. Remote Sens.,36, 206–213.

    • Crossref
    • Export Citation
  • ——, 1998b: Linkages between global vegetation and climate: An analysis based on NOAA-Advanced Very High Resolution Radiometer data. Ph.D. dissertation, Vrije Universiteit Amsterdam, 199 pp. [Also available as NASA Rep. GSFC/CR-1998-206852;NASA-CASI, Hanover, MD 21076-1320.].

  • ——, C. O. Justice, and C. J. Tucker, 1994: A global 1° by 1° NDVI dataset for climate studies derived from the GIMMS continental NDVI. Int. J. Remote Sens.,15, 3493–3518.

    • Crossref
    • Export Citation
  • Loveland, T. R., and A. S. Belward, 1997: The IGBP-DIS global 1 km landcover dataset, DISCover: First results. Int. J. Remote Sens.,18, 3289–3295.

    • Crossref
    • Export Citation
  • Malmström, C. M., M. V. Thompson, G. Juday, S. O. Los, J. T. Randerson, and C. B. Field, 1997: Interannual variation in global-scale primary production: Testing model estimates. Global Biochem. Cycles,11, 367–392.

    • Crossref
    • Export Citation
  • Myneni, R. B., S. O. Los, and C. J. Tucker, 1995: Satellite-based identification of linked vegetation index and sea surface temperature anomaly areas from 1982–1990 for Africa, Australia and South America. Geophys. Res. Lett.,23, 729–732.

    • Crossref
    • Export Citation
  • ——, R. R. Nemani, and S. W. Running, 1997a: Estimation of global leaf area index and absorbed PAR using radiative transfer models. IEEE Trans. Geosci. Remote Sens.,35, 1380–1393.

    • Crossref
    • Export Citation
  • ——, C. D. Keeling, C. J. Tucker, G. Asrar, and R. R. Nemani, 1997b:Increased plant growth in the northern high latitudes from 1981 to 1991. Nature,386, 698–702.

    • Crossref
    • Export Citation
  • Nemani, R. R., S. W. Running, R. A. Pielke, and T. N. Chase, 1996:Global vegetation cover changes from coarse resolution satellite data. J. Geophys. Res.,101, 7157–7162.

    • Crossref
    • Export Citation
  • Potter, C. S., J. T. Randerson, C. B. Field, P. A. Matson, P. M. Vitousek, H. A. Mooney, and S. A. Klooster, 1993: Terrestial ecosystem production: A process model based on global satellite and surface data. Global Geochem. Cycles,7, 811–841.

    • Crossref
    • Export Citation
  • Price, J. C., 1987: Calibration of satellite radiometers and the comparison of vegetation indices. Remote Sens. Environ.,21, 15–27.

    • Crossref
    • Export Citation
  • Privette, J. L., C. Fowler, G. A. Wick, D. Baldwin, and W. J. Emery, 1995: Effects of orbital drift on Advanced Very High Resolution Radiometer products: Normalized difference vegetation index and sea surface temperature. Remote Sens. Environ.,53, 164–171.

    • Crossref
    • Export Citation
  • Randall, D. A., and Coauthors, 1996: A revised land-surface parameterization (SiB2) for GCMs. Part III: The greening of the Colorado State University general circulation model. J. Climate,9, 738–763.

    • Crossref
    • Export Citation
  • Randerson, J. T., M. V. Thompson, C. M. Malmstrom, C. B. Field, and I. Y. Fung, 1996: Substrate limitations for heterotrophs: Implications for models that estimate the seasonal cycle of atmospheric CO2. Global Biochem. Cycles,10, 585–602.

  • Reyment, R., and K. G. Jöreskog, 1993: Applied Factor Analysis in the Natural Sciences. Cambridge University Press, 371 pp.

  • Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev.,115, 1606–1626.

    • Crossref
    • Export Citation
  • Roujean, J. L., M. Leroy, and P. Y. Deschamps, 1992: A bidirectional reflectance model of the earth’s surface for the correction of remote sensing data. J. Geophys. Res.,97, 20 455–20 468.

    • Crossref
    • Export Citation
  • Ruimy, A., B. Saugier, and G. Dedieu, 1994: Methodology for the estimation of terrestrial net primary production from remotely sensed data. J. Geophys. Res.,99, 5263–5283.

    • Crossref
    • Export Citation
  • Sellers, P. J., 1985: Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens.,6, 1335–1372.

    • Crossref
    • Export Citation
  • ——, J. A. Berry, G. J. Collatz, C. B. Field, and F. G. Hall, 1992: Canopy reflectance, photosynthesis and transpiration, III. A re-analysis using improved leaf models and a new canopy integration scheme. Remote Sens. Environ.,42, 187–216.

    • Crossref
    • Export Citation
  • ——, D. A. Randall, G. J. Collatz, J. A. Berry, C. B. Field, D. A. Dazlich, C. Zhang, and L. Bounoua, 1996a: A revised land-surface parameterization (SiB2) for GCMs. Part I: Model formulation. J. Climate,9, 676–705.

    • Crossref
    • Export Citation
  • ——, S. O. Los, C. J. Tucker, C. O. Justice, D. A. Dazlich, G. J. Collatz, and D. A. Randall, 1996b: A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate,9, 706–737.

    • Crossref
    • Export Citation
  • ——, and Coauthors, 1996c: The ISLSCP Initiative I global datasets:Surface boundary conditions and atmospheric forcings for land–atmosphere studies. Bull. Amer. Meteor. Soc.,77, 1987–2005.

    • Crossref
    • Export Citation
  • Singh, S. M., 1988: Simulation of solar zenith angle effect on global vegetation index (GVI) data. Int. J. Remote Sens.,9, 237–248.

    • Crossref
    • Export Citation
  • Strebel, D. E., D. R. Landis, K. F. Huemmrich, and B. W. Meeson, 1994: Collected data of the First ISLSCP Field Experiment, vol. 1: Surface observations and non-image data sets. CD-ROM, NASA. [Available from Oak Ridge National Laboratory P.O. Box 2008, DAAC, MS 6407, Oak Ridge, TN 37831.].

  • Tucker, C. J., and P. J. Sellers, 1986: Satellite remote sensing of primary production. Int. J. Remote Sens.,7, 1395–1416.

    • Crossref
    • Export Citation
  • ——, H. E. Dregne, and W. W. Newcomb, 1994: AVHRR datasets for determination of desert spatial extent. Int. J. Remote Sens.,15, 3547–3565.

    • Crossref
    • Export Citation
  • Vermote, E. F., N. El Saleous, Y. J. Kaufman, and E. Dutton, 1997a:Data preprocessing: Stratospheric aerosol perturbing effect on the remote sensing of vegetation: Correction method for the composite NDVI after the Pinatubo eruption. Remote Sens. Rev.,15, 7–21.

    • Crossref
    • Export Citation
  • ——, D. Tanré, J. L. Deuzé, M. Herman, and J. J. Morcette, 1997b:Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An overview. IEEE Trans. Geosci. Remote Sens.,35, 675–686.

    • Crossref
    • Export Citation

Fig. 1.
Fig. 1.

Interferences in NDVI data. (a) Variations in NDVI over the Saharan desert that result from calibration differences and sensitivity changes in AVHRRs on board NOAA-7, -9, and -11. (b) Effects of optical thickness from El Chichón aerosols on average monthly NDVI values >0.5. (c) Temporal changes and missing data from clouds in an NDVI time series over a tropical forest with otherwise low temporal variation. (d) Decreased NDVI values at the start of the rainy season in the Sahel. Deviations are related to high atmospheric water vapor in the ITCZ (Justice et al. 1991). Dotted lines indicate estimates of NDVI with water vapor effects removed. (e) (top line) Viewing angle–dependent variations in the Roujean et al. (1992) model fitted through NDVI data from FIFE (Deering and Leone 1986). (bottom line) Viewing angle–dependent variations in the same NDVI data at the top of the atmosphere. Viewing angle variations were simulated with 6S (Vermote et al. 1997b) for a midlatitude continental atmosphere with fairly high aerosol concentration. The ground-measured NDVI data and top-of-the-atmosphere data have an inverse relationship. A small hot-spot effect exists at 30° off nadir in the backscatter direction. (f) The effect of solar zenith angle on NDVI (ΔNDVI/Δθ) estimated from the global dataset (line marked 1) compared with ground-measured NDVI (top curve) and simulated top-of-the-atmosphere NDVI for clear (optical thickness = 0.1) and hazy (optical thickness = 0.5) atmospheres. The solar zenith angle effect for line “1” is within the simulations for clear and hazy atmospheres.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 2.
Fig. 2.

Outline of FASIR corrections: (a) Original NDVI time series with outliers (two lowest NDVI values). (b) NDVI time series after Fourier adjustment (FA of FASIR; open circles are new estimates). (c) Relationship between NDVI and solar zenith angle for fully green canopies. For a given solar zenith angle, its effect is assumed to be linear with NDVI. (d) Interpolation of missing data during the winter. (e) Reconstruction of NDVI time series over tropical forests. The seasonality of tropical forests is removed.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 3.
Fig. 3.

Evaluation of the Fourier adjustment. The months of Jan, Apr, Jul, and Oct were removed from a sequence of 12 months and their NDVI values were estimated with the Fourier adjustment. The figure shows the frequency distributions of the difference between the estimated and observed NDVI data with several of the statistics of the distributions (minimum, 5th percentile, median; 95th percentile, maximum). No comparison was made for missing values in the observed NDVI data.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 4.
Fig. 4.

Derivation of FPAR absorbed by the vegetation canopy and leaf area index from NDVI. (a) Calculation of 2d and 98th percentile values from NDVI frequency distributions per vegetation class. The 98th percentile values were derived from tall vegetation class (assumed to be fully green) and the 2d percentile values were derived from desert and bare soil and shrub classes (assumed to be bare). (b) Calculation of FPAR from NDVI. FPAR is estimated as the mean of a linear relationship between FPAR and NDVI and a linear relationship between FPAR and SR = (1 + NDVI)/(1 − NDVI). The 98th percentile NDVI and SR values correspond to FPAR = 0.95 and the 2d percentile NDVI and SR values correspond to FPAR = 0.001. (c) Vegetation cover fraction fV is estimated as the maximum FPAR over the year divided by the maximum FPAR for the class (0.95). (d) Calculation of leaf area index from FPAR. An exponential relationship is used for vegetation within the vegetation cover fraction.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 5.
Fig. 5.

Evaluation of NDVI FPAR relationships with data from FIFE (Strebel et al. 1994; Hall et al. 1992), OTTER (Angelici et al. 1991), BOREAS [data provided by C. Walthall, P. Rich, K. F. Huemmrich, and the BOREAS information system; see also Fournier et al. (1997), Chen et al. (1997), and Chen (1996)], and HAPEX—Sahel (Hanan et al. 1997). (a) Measured FPAR vs FPAR estimated with the NDVI–FPAR model. (b) Measured FPAR vs FPAR estimated with the NDVI–SR model. (c) Measured FPAR vs mean estimated FPAR from the NDVI–FPAR and SR–FPAR model (see text for discussion).

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 6.
Fig. 6.

Relationship between NDVI corrected for Rayleigh scattering and ozone absorption and uncorrected NDVI. Data were taken from a N–S transect through Africa from the Sahara, through the equatorial forest, and into southern Africa. Data are from NOAA-9 maximum NDVI composites of Mar 1995.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 7.
Fig. 7.

Evaluation of FPAR–leaf area index relationships. The symbols indicate ground-measured FPAR vs leaf area index. The solid lines show the exponential model and the dotted lines the linear model used to estimate leaf area index from FPAR. (a) Total (from both live and dead vegetation) FPAR vs total leaf area index measured at the FIFE field campaign (Strebel et al. 1994). (b) FPAR vs leaf area index from a BOREAS site (Chen and Cihlar 1996). The solid line shows the exponential model and the dotted line the linear model used to estimate leaf area index from FPAR. (c) FPAR vs leaf area index from a BOREAS site (data provided by P. Rich, K. F. Huemmrich, and the BOREAS information system). The data marked “N” indicate measurements from needleleaf vegetation (either black spruce or jack pine) and the data marked “A” indicate measurements from an aspen site. The solid line shows the exponential model to estimate leaf area index from FPAR for needleleaf vegetation, the dotted line the linear model for needleleaf vegetation, and the dashed line the exponential model for deciduous broadleaf vegetation. Note that the fit of the linear model is inadequate for both (b) and (c).

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 8.
Fig. 8.

The x axis: mean annual NDVI averaged per biome type; the y axis: temporal standard deviation of NDVI anomaly time series aggregated per biome. Numbers indicate the SiB2 land cover types (Table 1).

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 9.
Fig. 9.

Interannual variation for 1982–90 in crop yield data of selected sites and in annual global NPP (Gt carbon), based on CASA model runs using different NDVI versions. (a) Time series of residuals from regression between yield and NPP estimates from FASIR, Pathfinder, and GIMMS NDVI for a Turkish agricultural site. (b) Same, but for a Polish site. (c) Global estimates of NPP from FASIR, Pathfinder, and GIMMS data. (d) Optical thickness (τ) estimates at the equator from stratospheric aerosols by the El Chichón eruption and equatorial crossing time (ECT) of NOAA-7, -9, and -11. Analysis of variance estimating global NPP as a function of τ and ECT indicates significance levels of [Pr(τ) = 0.04, Pr(ECT) = 0.07] for global FASIR NPP, of [Pr(τ) = 0.007, Pr(ECT) = 0.005] for global Pathfinder NPP, and of [Pr(τ) = 0.02, Pr(ECT) = 0.0009] for global GIMMS NPP (data from Malmström et al. 1997).

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Fig. 10.
Fig. 10.

Interannual variation in departures from monthly mean leaf area index (LAI; thick continuous line), precipitation (boxes), and land surface air temperature (thin dotted line) averaged over selected parts of the growing season. Sea surface data are identified with the land cover classification map and were excluded. (a) Western Europe from 64°N, 3°W–46°N, 13°E for early spring (Mar, Apr, May). (b) Same but for summer (Jun–Sep). (c) Nordeste (Brazil) 3°S, 42°W–12°S, 44°W for the entire year. (d) Sahel 17°N, 7°W–12°N, 23°E for Jul–Sep.

Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0183:AGYBLS>2.0.CO;2

Table 1.

Vegetation cover types as defined in this paper with examples of the associated properties: maximum leaf area index LT,max, stem area index LS, NDVI at 2d and 98th percentiles of NDVI distributions, parameters for the solar zenith angle correction k1 through k4 [see Eq. (3)]. For the solar angle correction only, a distinction is made within SiB2 biome 6 between the old SiB1 biome 6, which is treated as morphologically similar to class 1, and SiB1 biomes 7, 8, and 11, which are treated as morphologically similar to the short vegetation classes.

Table 1.
Table 2.

Analysis of variance of alternative models to estimate FPAR from NDVI. Here Df is degrees of freedom, SS = sum of squares, and MS is mean squares. Model sum of squares and degrees of freedom and residual sum of squares and degrees of freedom (rows 1 and 2 of table) add up to total sum of squares and degrees of freedom (row 5). Lack-of-fit sum of squares and degrees of freedom (row 3) and pure-error sum of squares and degrees of freedom (row 4) add up to residual sum of squares (row 2). Mean of squares is calculated as the sum of squares divided by the degrees of freedom. F ratio is the ratio between the means of squares (see text for discussion).

Table 2.
Table 3.

Comparison of NPP estimates from the CASA model based on different NDVI versions. Shown are the fraction of observations with strong correlation r (P < 0.05) and in parentheses the fraction with strong correlation or moderate correlation (P < 0.10). The latter is included because the short length of the time series reduces statistical power. SZA stands for solar zenith angle.

Table 3.
Table 4.

Vegetation classes in which trend in net primary production estimates is evident, by magnitude of trend (% yr−1).

Table 4.
Table 5.

Loadings on the first principal component (Reyment and Jöreskog 1993) of scaled precipitation, land surface air temperature, and leaf area index anomalies for sites of Fig. 10 (see text for discussion).

Table 5.
Table 6.

Sensitivity of a selection of hydrometeorological land surface parameters to changes in vegetation in the SiB2 CSU GCM. Here Δ indicates change.

Table 6.
Save